feat: update kimi-code skill with session persistence, file arch refactor issues #164-#170

2026-03-15 11:38:17 -04:00
parent d785af7f2f
commit e1ba3fcf6b
5 changed files with 148 additions and 42 deletions
--- a/bin/timmy-loop.sh
+++ b/bin/timmy-loop.sh
@@ -1,7 +1,12 @@
 #!/usr/bin/env bash
-# ── Timmy Development Loop v2 ──────────────────────────────────────────
-# Runs forever. Each cycle: Timmy triages → Hermes picks work → execute
-# → Timmy reviews → merge. State in .loop/state.json.
+# ── Timmy Development Loop v3 ──────────────────────────────────────────
+# Runs forever. Each cycle: triage → Hermes executes → retro → repeat.
+#
+# TRIAGE (two tiers):
+#   Fast (every 5 cycles):  mechanical scoring → .loop/queue.json
+#   Deep (every 20 cycles): Hermes reads code + consults Timmy → refined queue
+#
+# RETRO (every cycle): structured log → .loop/retro/cycles.jsonl
 # ───────────────────────────────────────────────────────────────────────

 set -uo pipefail
@@ -17,6 +22,12 @@ LOCKFILE="/tmp/timmy-loop.lock"
 COOLDOWN=3
 MAX_CYCLE_TIME=1200          # 20 min — enough for complex issues
 CLAIM_TTL_SECONDS=3600       # 1 hour — stale claims auto-expire
+FAST_TRIAGE_INTERVAL=5       # mechanical scoring every N cycles
+DEEP_TRIAGE_INTERVAL=20      # Hermes+Timmy deep triage every N cycles
+TRIAGE_SCRIPT="$REPO/scripts/triage_score.py"
+RETRO_SCRIPT="$REPO/scripts/cycle_retro.py"
+DEEP_TRIAGE_SCRIPT="$REPO/scripts/deep_triage.sh"
+QUEUE_FILE="$REPO/.loop/queue.json"
 # macOS doesn't have timeout; use perl fallback
 if ! command -v timeout &>/dev/null; then
    timeout() {
@@ -125,11 +136,69 @@ except Exception as e:
    print(f'[cleanup] Claim cleanup failed: {e}')
 " 2>&1
 }
+# ── Fast triage (mechanical scoring) ──────────────────────────────────
+run_fast_triage() {
+    if [ -f "$TRIAGE_SCRIPT" ]; then
+        log "Running fast triage..."
+        python3 "$TRIAGE_SCRIPT" 2>&1
+    else
+        log "WARNING: triage script not found at $TRIAGE_SCRIPT"
+    fi
+}

+# ── Deep triage (Hermes + Timmy) ─────────────────────────────────────
+run_deep_triage() {
+    if [ -f "$DEEP_TRIAGE_SCRIPT" ]; then
+        log "Running deep triage (Hermes + Timmy)..."
+        timeout 600 bash "$DEEP_TRIAGE_SCRIPT" 2>&1
+    else
+        log "WARNING: deep triage script not found at $DEEP_TRIAGE_SCRIPT"
+    fi
+}

+# ── Cycle retrospective ─────────────────────────────────────────────
+log_retro() {
+    # Usage: log_retro <success|failure> [extra args for cycle_retro.py]
+    local outcome="$1"; shift
+    if [ -f "$RETRO_SCRIPT" ]; then
+        local duration=$(( $(date +%s) - CYCLE_START ))
+        if [ "$outcome" = "success" ]; then
+            python3 "$RETRO_SCRIPT" --cycle "$CYCLE" --success \
+                --duration "$duration" "$@" 2>&1
+        else
+            python3 "$RETRO_SCRIPT" --cycle "$CYCLE" --failure \
+                --duration "$duration" "$@" 2>&1
+        fi
+    fi
+}
+
+# ── Inject queue context into prompt ─────────────────────────────────
+get_queue_context() {
+    if [ -f "$QUEUE_FILE" ]; then
+        python3 -c "
+import json
+with open('$QUEUE_FILE') as f: q = json.load(f)
+if not q:
+    print('Queue is empty. Check open issues on Gitea and pick the highest-priority one.')
+else:
+    print(f'PRIORITIZED QUEUE ({len(q)} ready issues):')
+    for i, item in enumerate(q[:8]):
+        flag = 'BUG' if item.get('type') == 'bug' else item.get('type', '?').upper()
+        print(f'  {i+1}. #{item[\"issue\"]} [{flag}] score={item[\"score\"]} — {item.get(\"title\",\"?\")[:60]}')
+        if item.get('files'):
+            print(f'     files: {\", \".join(item[\"files\"][:3])}')
+    if len(q) > 8:
+        print(f'  ... +{len(q)-8} more')
+    print()
+    print(f'Pick from the TOP of this queue. Issue #{q[0][\"issue\"]} is highest priority.')
+" 2>/dev/null
+    else
+        echo "No queue file. Check open issues on Gitea and pick the highest-priority one."
+    fi
+}

 # ── Main Loop ─────────────────────────────────────────────────────────
-log "Timmy development loop v2 starting. PID $$"
+log "Timmy development loop v3 starting. PID $$"
 log "Timeout: ${MAX_CYCLE_TIME}s | Cooldown: ${COOLDOWN}s | Claim TTL: ${CLAIM_TTL_SECONDS}s"
 log "Repo:  $REPO"
 update_state "started_at" "\"$(date -u +%Y-%m-%dT%H:%M:%SZ)\""
@@ -152,13 +221,25 @@ while true; do
    update_state "cycle" "$CYCLE"
    update_state "status" '"working"'

+    CYCLE_START=$(date +%s)
+
    # ── Pre-cycle housekeeping ────────────────────────────────────────
    expire_claims

-    # ── Build the prompt with time budget ──────────────────────────────
+    # ── Triage (fast every 5, deep every 20) ─────────────────────────
+    if (( CYCLE % DEEP_TRIAGE_INTERVAL == 0 )); then
+        run_deep_triage
+    elif (( CYCLE % FAST_TRIAGE_INTERVAL == 0 )); then
+        run_fast_triage
+    fi
+
+    # ── Build the prompt with time budget + queue ────────────────────
+    QUEUE_CONTEXT=$(get_queue_context)
    PROMPT=$(cat "$PROMPT_FILE")
    PROMPT="TIME BUDGET: You have $((MAX_CYCLE_TIME / 60)) minutes for this cycle. Plan accordingly — do not start work you cannot finish.

+$QUEUE_CONTEXT
+
 $PROMPT"

    log "Spawning hermes for cycle $CYCLE..."
@@ -169,6 +250,8 @@ $PROMPT"
        update_state "status" '"idle"'
        update_state "last_completed" "\"$(date -u +%Y-%m-%dT%H:%M:%SZ)\""

+        # ── Cycle retro (success) ────────────────────────────────────
+        log_retro success

    else
        EXIT_CODE=$?
@@ -182,7 +265,10 @@ errs.append({'cycle': $CYCLE, 'code': $EXIT_CODE, 'time': '$(date -u +%Y-%m-%dT%
 s['errors'] = errs[-10:]
 with open('$STATE', 'w') as f: json.dump(s, f, indent=2)
 "
-        # ── Cleanup on failure ────────────────────────────────────────
+        # ── Cycle retro (failure) ────────────────────────────────────
+        log_retro failure --reason "exit code $EXIT_CODE"
+
+        # ── Cleanup on failure ───────────────────────────────────────
        cleanup_cycle "$CYCLE"
    fi

--- a/bin/timmy-tmux.sh
+++ b/bin/timmy-tmux.sh
@@ -1,14 +1,14 @@
 #!/usr/bin/env bash
 # ── Timmy Loop tmux Session ────────────────────────────────────────────
-# Creates session with 3 panes using standard tmux splits.
+# Creates session with 4 panes.
 #
-# Layout:
-#   ┌──────────────────────┬──────────────────────┐
-#   │    LOOP (small)      │                      │
-#   ├──────────────────────┤   HERMES CHAT        │
-#   │    STATUS DASHBOARD  │   (full height)      │
-#   │    (live refresh)    │                      │
-#   └──────────────────────┴──────────────────────┘
+# Layout (245×62 terminal):
+#   ┌────────────────────────────────────────────┬──────────────────────┐
+#   │              LOOP (10 rows)                │                      │
+#   ├──────────────────────┬─────────────────────┤   CHAT (full height) │
+#   │    STATUS (81 cols)  │  LOOPSTAT (40 cols) │                      │
+#   │    (50 rows)         │  (50 rows)          │                      │
+#   └──────────────────────┴─────────────────────┴──────────────────────┘
 # ───────────────────────────────────────────────────────────────────────

 SESSION="timmy-loop"
@@ -21,22 +21,26 @@ sleep 1
 # Create session — pane 0 starts as shell
 tmux new-session -d -s "$SESSION" -x 245 -y 62

-# Vertical split: left (50%) | right (50%)
+# Vertical split: left (~50%) | right Chat (~50%)
 tmux split-window -h -p 50 -t "$SESSION:0.0"

-# Horizontal split on left pane: Loop (small top) / Status (big bottom)
-# Loop gets ~14% height (10 rows out of ~62), Status gets the rest
+# Horizontal split on left: Loop (small top ~16%) | bottom (~84%)
 tmux split-window -v -p 83 -t "$SESSION:0.0"

+# Vertical split on bottom-left: Status (wide ~67%) | LOOPSTAT (~33%)
+tmux split-window -h -p 33 -t "$SESSION:0.1"
+
 # Pane map after splits:
-#   0 = top-left (small)   → Loop output
-#   1 = bottom-left (big)  → Status dashboard
-#   2 = right (full height) → Hermes chat
+#   0 = top-left (full width) → Loop output
+#   1 = bottom-left (wide)    → Status dashboard
+#   2 = bottom-mid (narrow)   → LOOPSTAT (strategy)
+#   3 = right (full height)   → Hermes chat

 # Set titles
 tmux select-pane -t "$SESSION:0.0" -T "Loop"
 tmux select-pane -t "$SESSION:0.1" -T "Status"
-tmux select-pane -t "$SESSION:0.2" -T "Chat"
+tmux select-pane -t "$SESSION:0.2" -T "LOOPSTAT"
+tmux select-pane -t "$SESSION:0.3" -T "Chat"

 # Pane border styling
 tmux set-option -t "$SESSION" pane-border-status top
@@ -47,17 +51,18 @@ tmux set-option -t "$SESSION" pane-active-border-style "fg=cyan"
 # Start processes
 tmux send-keys -t "$SESSION:0.0" "export PATH=\"$HOME/.local/bin:$HOME/.hermes/bin:/usr/local/bin:\$PATH\" && $HOME/.hermes/bin/timmy-loop.sh" Enter
 tmux send-keys -t "$SESSION:0.1" "$HOME/.hermes/bin/timmy-status.sh" Enter
-tmux send-keys -t "$SESSION:0.2" "cd ~/Timmy-Time-dashboard && hermes" Enter
+tmux send-keys -t "$SESSION:0.2" "$HOME/.hermes/bin/timmy-loopstat.sh" Enter
+tmux send-keys -t "$SESSION:0.3" "cd ~/Timmy-Time-dashboard && hermes" Enter

 # Focus chat pane
-tmux select-pane -t "$SESSION:0.2"
+tmux select-pane -t "$SESSION:0.3"

 echo ""
-echo "  ┌──────────────────┬──────────────────┐"
+echo "  ┌──────────────────────────────────┬──────────────────┐"
 echo "  │         Loop (pane 0)            │                  │"
-echo "  ├──────────────────┤  Chat (pane 2)   │"
-echo "  │  Status (pane 1) │                  │"
-echo "  └──────────────────┴──────────────────┘"
+echo "  ├────────────────────┬─────────────┤  Chat (pane 3)   │"
+echo "  │  Status (pane 1)   │ LOOPSTAT(2) │                  │"
+echo "  └────────────────────┴─────────────┴──────────────────┘"
 echo ""
 echo "  Attach:  tmux attach -t timmy-loop"
 echo "  Stop:    touch ~/Timmy-Time-dashboard/.loop/STOP"
--- a/cron/jobs.json
+++ b/cron/jobs.json
@@ -35,12 +35,12 @@
      "schedule_display": "every 8m",
      "repeat": {
        "times": null,
-        "completed": 20
+        "completed": 28
      },
      "enabled": true,
      "created_at": "2026-03-14T21:25:31.831983-04:00",
-      "next_run_at": "2026-03-15T10:17:06.096722-04:00",
-      "last_run_at": "2026-03-15T10:09:06.096722-04:00",
+      "next_run_at": "2026-03-15T11:35:42.310507-04:00",
+      "last_run_at": "2026-03-15T11:27:42.310507-04:00",
      "last_status": "ok",
      "last_error": null,
      "deliver": "local",
@@ -58,12 +58,12 @@
      "schedule_display": "every 8m",
      "repeat": {
        "times": null,
-        "completed": 18
+        "completed": 26
      },
      "enabled": true,
      "created_at": "2026-03-14T21:39:34.712372-04:00",
-      "next_run_at": "2026-03-15T10:17:32.633122-04:00",
-      "last_run_at": "2026-03-15T10:09:32.633122-04:00",
+      "next_run_at": "2026-03-15T11:35:59.953250-04:00",
+      "last_run_at": "2026-03-15T11:27:59.953250-04:00",
      "last_status": "ok",
      "last_error": null,
      "deliver": "local",
@@ -81,17 +81,17 @@
      "schedule_display": "every 10m",
      "repeat": {
        "times": null,
-        "completed": 3
+        "completed": 8
      },
      "enabled": true,
      "created_at": "2026-03-15T09:23:09.042883-04:00",
-      "next_run_at": "2026-03-15T10:16:56.086710-04:00",
-      "last_run_at": "2026-03-15T10:06:56.086710-04:00",
+      "next_run_at": "2026-03-15T11:32:58.861581-04:00",
+      "last_run_at": "2026-03-15T11:22:58.861581-04:00",
      "last_status": "ok",
      "last_error": null,
      "deliver": "local",
      "origin": null
    }
  ],
-  "updated_at": "2026-03-15T10:09:32.633259-04:00"
+  "updated_at": "2026-03-15T11:27:59.953394-04:00"
 }
--- a/memories/MEMORY.md
+++ b/memories/MEMORY.md
@@ -1,8 +1,8 @@
-Gitea (localhost:3000): Users: rockachopa(admin), hermes(id=4), manus(id=3), kimi(id=5). Repos: rockachopa/Timmy-time-dashboard, rockachopa/hermes-agent, rockachopa/hermes-config. Hermes token: ~/.hermes/gitea_token. DO NOT use Alex's token — ever. Future work: agentd user isolation.
+Gitea (localhost:3000): Users: rockachopa(admin), hermes(id=4), manus(id=3), kimi(id=5). Repos: rockachopa/Timmy-time-dashboard, hermes/hermes-config (private, rockachopa=admin collab). Hermes token: ~/.hermes/gitea_token. DO NOT use Alex's token — ever. Sync tool: ~/.hermes/bin/hermes-config-sync — run after modifying config/scripts/memories/skills. Future work: agentd user isolation.
 §
 Timmy architecture plan: Replace hardcoded _PERSONAS and TimmyOrchestrator with YAML-driven agent config (agents.yaml). One seed class, all differentiation via config. User wants to update YAML files, not Python, to add capabilities. Key files to refactor: agents/timmy.py, agents/base.py, tools_delegation/__init__.py, tools_intro/__init__.py. SubAgent class stays mostly as-is, just reads from YAML. Routing should be config-based pattern matching, not LLM calls. Per-agent model assignment (big model for orchestrator/code/research, small for simple tasks). qwen3:30b pulling as primary local model (~18GB, MoE 3B active).
 §
-Hermes-Timmy workspace: ~/Timmy-Time-dashboard/workspace/. Flat file correspondence, append-only.
+Kimi delegation: 1-3 files max, pre-extract context (scans full codebase otherwise), ~/worktrees/ not /tmp/, verify commits, two-attempt rule. Worktrees: ~/worktrees/kimi-*.
 §
 2026-03-14: Issues #36-#40,#52 fixed. Voice loop, fallback chain, self-prompt queue built. Timmy on qwen3:30b (4096 ctx). Loop v2 with 20min timeout. Filed eval issues #77-#87. Status: ~/.hermes/bin/timmy-status.sh.
 §
--- a/skills/autonomous-ai-agents/kimi-code/SKILL.md
+++ b/skills/autonomous-ai-agents/kimi-code/SKILL.md
@@ -21,7 +21,22 @@ Delegate coding tasks to Kimi Code CLI via the Hermes terminal. Powered by kimi-
 - Logged in: `kimi login`
 - Config at `~/.kimi/config.toml`

-## One-Shot Tasks (Print Mode)
+## Session Persistence (ALWAYS USE)
+
+Kimi scans the entire codebase on first run. Reuse sessions to keep context warm:
+
+```
+# First task in a worktree — creates session, caches codebase
+terminal(command="kimi --print --session worktree-148 -p 'Fix the XSS vulnerability'", workdir="~/worktrees/kimi-148", timeout=300)
+
+# Subsequent tasks — context is cached, much faster
+terminal(command="kimi --print --continue --session worktree-148 -p 'Now fix the CSRF issue'", workdir="~/worktrees/kimi-148", timeout=180)
+```
+
+**NEVER run Kimi without --session on large codebases.** First run takes 2-5 min.
+Subsequent runs in same session take 30-90s.
+
+## One-Shot Tasks (only for tiny repos or quick questions)

 ```
 terminal(command="kimi --print -p 'Fix the XSS vulnerability in swarm_live.html by sanitizing innerHTML'", workdir="~/project", timeout=120)